Trash or Treasure: How to Utilize Emojis in Social Media Sentiment Classification


Summer Research Project Supervised by Prof. Mathieu Laurière. Funded by NYU Shanghai Dean’s Undergraduate Research Fund.

Emojis are usually “trashed out” during preprocessing stage of Twitter sentiment analysis. In this research, I probed into how we can incorporate those sentiment-rich emojis to improve sentiment analysis accuracy. I also conducted an experiment on how compatible current BERT-based encoders are with emojis. This work provides insights into how we should process emoji-included data with BERT encoders for the sentiment analysis task.

[Code] [Report] [TDS Blog]